Post Go back to editing

Factory DPD characterization / open loop coefficient load and operation possible?

Category: Hardware
Product Number: ADRV9002

Is it possible to build a LUT of DPD coefficients at the [customer] factory in-circuit prior to deployment across the parameters of temperature, power-level, and frequency that could be applied open loop in the field based on those known parameters, even at some compromised performance vs. real-time DPD feedback?

If so, what guidance can you offer on the size of the LUT? - i.e.. what resolution might be acceptable for measuring those three variables?

In the application of interest, an LDMOS PA would be used at up to ~10W, the temperature range would pretty much span the gamut, but the frequency range would be low S-band, only about a 10-15% range.

  • As I played with TES and the software allows users to save and update the coefficients on the fly, but upon examining the API function calls, they can be done only in CALIBRATED state. It means that you need to pause transmission of data, change the radio state from RF_ENABLED to CALIBRATED, save/load DPD coefficients, and change the state back to RF_ENABLED to resume transmission. Using this mechanism, you can train DPD in various conditions and save the coefficients into files, and reload the desired coefficients when environment changes. Wondering whether this will meet your requirements. Currently only 256 or 512 of LUT size is supported, we recommend to start with default size that you can get from TES.

    Hope this helps, and let us know if you need any further clarification.

    -YH

  • ---- edited from original posting ---

    Thanks YH.

    I see in UG-1828 that the table "Time Taken to Change State" suggests 25.1 us each for migration of states from RF_Enabled to Calibrated and the converse.  So I envision this process taking 50.2 us plus the time to load new coefficients.  Is that right?

    I see that I have not have been clear enough about the LUT, though - probably because the idea of having a LUT for DPD is new to me, whereas I'm already familiar with the idea of using a LUT related to frequency hopping - e.g. to maintain QEC and LOL performance. 
    You advised a limitation around the size of LUT that is supported.  I think that is a HW limitation related to DPD coefficients.  I'm hoping to figure out the overall external LUT size where the environment variables we need to accommodate in the LUT are (1) temperature, (2) Tx power level, (3) Tx frequency....... and the coefficients need to take care of LOL, QEC, DPD, and whatever other calibration info is accommodated to get best performance in an environment where the chip has been powered down, and suddenly it needs to be operating as quickly as possible, with no time to calibrate.  I imagine this could be a lot of information.

    Another way to ask the question is:  "What granularity is needed for those 3 variables (temp, power, freq) in order to build LUTs that will give both good QEC and LOL performance, but also significant DPD benefit?".  Multiply those granularities across their respective ranges needed, and then by the size of a single LUT that specifies all the coefficients needed, and that should tell us the total external LUT size that would be needed.

    Hope this clarifies the underlying objective? 

    I'm a bit concerned that the time required to go offline to load a new coefficient LUT could be a problem.  We'll have to check that with the customer - but it also loops back into the question of the external LUT size so we can estimate how long it will take to load those coefficients. 

    ~BE

  • Hi Brian,

    I understand your requirements. I don't think we have those data to determine the granularity and also probably we'll have to measure the time to load LUT, but let me check with the team and will get back to you.

    -YH